Research Of The Alignment Between Features Of Space Relationships In 2D Images And Describing Words

As we all know, perception of spatial relationships is a basic human cognitive ability. How to automatically construct the spatial relationships description of objects between spatial relations has become an important research field, one of which can be widely used in graphic and text converter, image retrieval, human-computer interaction, geographic information systems and other fields of study. In the above research, the spatial relations described in the model build, improve and systematize have important theoretical significance and practical value.The study in this paper is part of the model described spatial relations, it describes the results of the model to start, but it is also provided for the description of the spatial relationship of language and vocabulary support. This paper was mainly studied on how to associate the visual features of the image spatial relationships and its described terms. Based on the description model spatial relations, we can get the correspondence relations for spatial categories and vocabulary characteristic of the correspondence. So the key point of 2D image features and their spatial relationships described in terms of alignment is the sentence to determine its own vocabulary category, which is also a word classification problem. As the spatial relationship of the sentence characteristics, this paper describes the relationship for the spatial orientation of words based on the sentence pattern classification method. And the methods of the experiment are finished, which show that the method can classify most artificial described statement. After the classification, in order to enhance the richness of the corpus and considering the characters for the position of the word-formation, the result of segmentation in the "word pieces" model is proposed based on word-word fragment identification method, which can not only identify some of the un-logged direction word, but also can help to remove the non-words in study. Then, for a small part of the sentence pattern can not be classified in the words, we classified them by calculating the similarity from basic vocabulary words in the method of classification. And in the basis of alignment we continue fine-grained alignment, make alignment to word space in the main direction of orientation, direction and extent of the second qualifier extracted with features.The work in this paper does not only provide a rich training corpus for the spatial relationship between the model, but also contributed compositions by classification and sentence fragments identified as 2D images summary of spatial relations to express the common position of the word sentence.